Bank of Canada Banque du Canada

Working Paper 95-2 / Document de travail 95-2

ESTIMATING AND PROJECTING POTENTIAL OUTPUTUSING STRUCTURAL VAR METHODOLOGY:

The Case of the Mexican Economy

by

Alain DeSerres, Alain Guayand Pierre St-Amant

March 1995

ESTIMATING AND PROJECTING POTENTIAL OUTPUTUSING STRUCTURAL VAR METHODOLOGY:

The Case of the Mexican Economy

by

Alain DeSerres, Alain Guay and Pierre St-Amant

International DepartmentBank of Canada

Ottawa, Ontario, CanadaK1A 0G9

Tel.: (613) 782-7657Fax: (613) 782-7658

Internet: aguay@bank-banque-canada.ca

The views expressed in this study are those of the authors and do not necessarilyrepresent those of the Bank of Canada. Correspondence concerning this papershould be addressed to Alain Guay.

Acknowledgments

We would like to thank René Lalonde for useful comments and discussions, and

for his contribution to the development of some of the computer programs that we

used. Thanks are also due to John Murray, Robert Lafrance, Simon van Norden,

James Powell, Graydon Paulin and Ianthi Vayid for their comments and

suggestions, and to Robert Vigfusson for his excellent research assistance.

ISSN 1192-5434 ISBN 0-662-23123-6

Printed in Canada on recycled paper

Contents

Abstract / Résumé .............................................................................................v

1 Introduction...................................................................................................1

2 The methodology ...........................................................................................3

3 The data.........................................................................................................8

4 Results ...........................................................................................................9 4.1 Variance decomposition .......................................................................10 4.2 Impulse responses................................................................................11 4.3 The components of output and potential output ................................12 4.4 Projecting potential output..................................................................15

5 Conclusions..................................................................................................16

Appendix 1: Charts of the series ....................................................................19

Appendix 2: Unit-root and cointegration tests ..............................................21

Bibliography ....................................................................................................23

v

Abstract

In this paper the authors show how potential output can be estimated and

projected through an approach derived from the structural vector autoregression

methodology. This approach is applied to the Mexican economy. To identify

demand, supply and world oil shocks, the authors assume that demand shocks do

not have a permanent effect on output and that the international price of oil is

exogenous to the Mexican economy in the long term. They then calculate potential

output by adding the world oil and supply components to the drift in output. They

find that world oil shocks have been an important source of both actual and

potential output fluctuations over a sample period extending from 1965 to 1994.

However, they also find occurrences of important gaps between actual and

potential output.

Résumé

Dans la présente étude, les auteurs montrent qu’il est possible d’estimer et de

prévoir le niveau de production potentielle en utilisant une approche dérivée de la

méthode structurelle d’autorégression vectorielle. Cette approche est appliquée

dans la modélisation de l’économie mexicaine. Pour identifier les chocs de

demande et d’offre et les chocs de prix mondiaux du pétrole, les auteurs font

l’hypothèse que les chocs de demande n’ont pas d’effet permanent sur la

production et que les cours mondiaux du pétrole sont exogènes à l’économie

mexicaine à long terme. Ils calculent la production potentielle en ajoutant les

composantes des prix pétroliers et de l’offre à la dérive de la production. Les

auteurs constatent que les chocs de prix mondiaux du pétrole ont été une cause

importante de variations de la production observée et potentielle au cours de la

période 1965-1994, mais découvrent également d’importants écarts entre la

production observée et potentielle.

1

1 Introduction

Most macroeconomic models used for forecasting and policy analysis require

estimates of potential output. In these models, potential output represents the

level to which real GDP reverts as the effect of demand disturbances dissipates.1

The gap between actual and potential output is a key variable determining the

evolution of prices and wages. While output in excess of potential leads to higher

inflation, sustained disinflation requires output to fall below potential, ceteris

paribus. It is therefore essential to use an appropriate method to measure

potential output.2

In this paper, we show how potential output can be estimated and

projected through an approach derived from the structural vector autoregression

(SVAR) methodology developed by Shapiro and Watson (1988), Blanchard and

Quah (1989), and King et al. (1991).3 This methodology involves the estimation of

a vector autoregression (VAR) model for the particular economy under study. We

then identify different types of shocks by making long-term assumptions based on

macroeconomic theory. This approach is applied here to the Mexican economy.4 To

identify aggregate demand, aggregate supply and world oil shocks, we assume

that demand shocks do not have a permanent effect on output and that the

international price of oil is exogenous to the Mexican economy in the long run. We

then calculate potential output by adding the world oil and supply components to

the deterministic trend in output.

Even though both oil shocks and aggregate supply shocks have

permanent effects on the level of production, we chose to identify them separately

1. Demand disturbances are defined later.2. For a discussion of how the estimation of potential output can affect the formulation of

monetary policy, see Boschen and Mills (1990).3. See also Blanchard and Watson (1986) or Sims (1986) for a similar approach using short-

term instead of long-term restrictions.4. The choice of country was motivated by ongoing work to fully incorporate Mexico into the

Bank of Canada’s quarterly economic projection.

2

and to interpret them differently, mainly to reflect the historical predominance of

the oil production sector in Mexico. In particular, despite negative intermediate

input effects felt in many sectors of the Mexican economy, a permanent increase

in the international price of oil leads to a net expansion in total domestic output

in Mexico. For that reason, we refer to aggregate supply shocks as reflecting

productivity or labour supply shocks, and we refer to world oil shocks as reflecting

a permanent change in either supply or demand conditions in the international oil

market.

The SVAR approach has many advantages over other popular

methods of estimating potential output, such as trend-based methods, filter-based

methods (using, for example, the Hodrick-Prescott filter),5 the Beveridge-Nelson

decomposition (either univariate or multivariate),6 or the methodology proposed

by Kuttner (1994).

First, unlike these alternative methods, the components of output

that the SVAR approach identifies can be given an economic interpretation. For

example, we can interpret fluctuations in potential output as being caused by

certain types of shocks (with a degree of uncertainty), whereas the other methods

cannot.

Second, the method that we propose takes into account the short-

term dynamics of shocks on the permanent component of output, which, we argue,

should be part of potential output.7

Third, contrary to other methods, such as those based on the Hodrick-

Prescott filter, the proposed method does not require the imposition of an arbitrary

smoothing parameter.

5. See Laxton and Tetlow (1992).6. See Evans and Reichlin (1994).7. This is discussed in Section 2, The methodology.

3

Fourth, since our approach is based on the estimation of a statistical

model (unlike filter-based methods), confidence intervals can be derived, which

gives a measure of the uncertainty surrounding the estimates of the output gap

and potential output.

Finally, this method should be attractive to policy makers in that it

provides a structural interpretation of the latest output data, based on

information that is available when economic policy decisions have to be made

(contrary to two-sided filters, which use ex post data).

The rest of the paper is organized as follows. In Section 2, we present

the methodology. Section 3 describes the data. Section 4 presents our results and

illustrates how the model can be used to project potential output. We conclude and

discuss avenues for future research in Section 5.

2 The methodology

In order to distinguish among various sources of output fluctuation, we apply a

variant of the structural VAR methodology to an autoregressive system composed

of three variables. It is assumed that the rate of growth of output (y), the price of

oil (o), and a monetary aggregate (m) each follow a stationary stochastic process

that responds to three types of non-autocorrelated orthogonal shocks: supply

shocks (εs), world oil shocks (εo), and demand shocks (εd).8

The structural model can be given a moving-average representation

as follows:

8. Note that a four-variable VAR including prices gave similar results. We chose to presentthis three-variable VAR because the presence of obvious non-linearities in Mexico’s priceseries would have unnecessarily complicated the analysis.

4

(1)

where

and

To simplify, the variance of the structural shocks is normalized so

that , the identity matrix.

To identify this structural model, the autoregressive reduced-form

VAR of the model is first estimated:

(2)

where et is a vector of estimated residuals, q is the number of lags, and

.

Given that the stochastic process is stationary, (2) may be written as

an infinite moving-average process:

(3)

The residuals of the model’s reduced form are thus related to the

structural residuals in the following way:

(4)

∆xt A0εt A1εt 1− …+ + Aiεt i−i 0=

∞

∑ A L( ) εt= = =

εt

εsεoεd

= ∆xt

∆y∆o∆m

=

E εtεt( ) I=

∆xt Π1∆xt 1− … Πq∆xt q− et+ + +=

E etet( ) Σ=

∆xt et C1et 1− …+ + Ciet i−i 0=

∞

∑ C L( ) et= = =

et A0εt=

5

which implies that

(5)

and thus,

(6)

In order to identify the structural shocks (ε) from the information

obtained by estimating the VAR (equation 2), that is, from the reduced-form

shocks (e) and their variance (Σ), we need to provide sufficient identifying

restrictions to evaluate the elements in A0. In this three-variable system, A0 has

nine elements. Since the estimated variance-covariance matrix Σ is symmetric,

equation (6) provides six independent identifying restrictions. Thus, three

additional restrictions must be imposed. From (1), (4) and (6), we note that the

matrix of long-run effects of reduced-form shocks, that is, C(1), is related to the

equivalent matrix of structural shocks, that is , through the following

relation:

(5)

where the matrix C(1) is calculated from the estimated VARs.9

If the variables considered in this paper were cointegrated, this, in

itself, would imply certain long-run restrictions.10 However, the results of

cointegration tests presented in Section 3 suggest that cointegration is not

present. Therefore, three identifying restrictions are imposed on A(1), based on

economic theory. One of these follows from the hypothesis that demand shocks

have no permanent effects on output. This implies that all shocks that have a

transitory effect on the level of output are interpreted as demand shocks. This

9. C(1) is the polynomial value for .10. On this subject, see King et al. (1991).

E etet( ) A0E εtεt( ) A0′=

A0A0′ Σ=

A L( )

A 1( ) C 1( ) A0=

C L( ) L 1=

6

rests on the assumption also used by Blanchard and Quah and others, that in the

long run output depends on factors such as productivity, demographics and

permanent shifts in the terms of trade. The second and third restrictions, which

allow us to distinguish between world oil shocks and supply shocks, are derived

from the assumption that only the former have a long-run effect on the price of

oil. Put another way, Mexico is assumed to be a price-taker on the world oil

market.

World oil shocks can have a permanent effect on the price of oil, since

this series follows a nonstationary process. The identified oil shocks could be

demand or supply shocks on the world market. Since demand for Mexican oil is

assumed to be perfectly elastic, a permanent increase in the price of oil can bring

a permanent increase in Mexican oil production. Given the importance of the oil

sector in Mexico, world oil shocks are expected to have a permanent effect on the

output level.

Therefore, we obtain the following structural output decomposition:

(6)

The right-hand side of equation (7), which is composed of the

moving-average components of the different types of shocks plus the

deterministic trend in output ( ), can be rewritten to take into account the

cyclical components of the supply and world oil shocks:

(7)

where the represent the transitory components of the permanent (supply)

shocks. The first five terms on the right-hand side of (8) represent our measure of

the first difference of potential output. We can then project potential output by

∆xt µ As L( ) εst Ao L( ) εot Ad L( ) εdt+ + +=

µ

∆xt µ As 1( ) εst As* L( ) εst Ao 1( ) εot Ao

* L( ) εot Ad L( ) εdt+ + + + +=

A* L( )

7

adding the effect of past oil and supply shocks to the projected deterministic

trend in output. Confidence intervals around the estimated potential output

series are obtained following a standard bootstrap approach.11

One difference between our approach and those of Kuttner (1994)

and Evans and Reichlin (1994) (based on Beveridge-Nelson decompositions) is in

the treatment of the transitory components of permanent shocks . Kuttner

identifies potential output as a random walk that is orthogonal to the cyclical

component of output. The decomposition he obtains may be represented in the

following way:

(8)

where xt includes output and prices, and potential output is defined as the

deterministic trend plus the permanent component (first two terms on the right-

hand side), which is orthogonal to the transitory component.

In the case of the Beveridge-Nelson method (univariate or

multivariate), the permanent and cyclical components are perfectly correlated.

Their decomposition can be expressed in the following way:

(9)

In this context, the first difference of potential output is simply the first two

terms on the right-hand side of the equation (10). Potential output is perfectly

correlated with the cyclical component, which is the last term of the equation.12

Unlike our approach, the Kuttner and Beveridge-Nelson approaches

11. For a survey of the bootstrap method see Jeong and Maddala (1993). For an application seeRunkle (1987).

12. It is interesting to note that the discussion of equations (8) to (10) can be seen as anextension in a multivariate context of the discussion contained in Watson (1986).

A* L( )

∆xt µ Γ 1( ) η1t Γ L( ) η2t+ +=

∆xt µ C 1( ) εt C* L( ) εt+ +=

8

do not take into account the transitory component of potential output. Since real-

world supply shocks are likely to be characterized by such a transitory component,

(for example, it is likely that the effect of a productivity shock will be felt only

gradually) we think that this should be seen as an advantage of our approach.13

Note that it is very important that the estimated VAR include a

sufficient number of lags. Monte Carlo simulations done by DeSerres and Guay

(1995) show that using a lag structure that is too parsimonious can significantly

bias the estimation of the structural components. These authors also find that

information-based criteria, such as the Akaike and Schwarz criteria, tend to select

an insufficient number of lags, while Wald or likelihood ratio (LR) tests, applied

according to a general-to-specific strategy, perform much better. Accordingly, we

selected the number of lags (12) to be included in our VAR on the basis of such an

LR test (using a 5 per cent critical value).

3 The data

We use quarterly data on industrial production and the monetary base (excluding

reserves at the central bank) taken from International Financial Statistics

(International Monetary Fund). Industrial production data, rather than real

GDP data, are used, as there is no appropriate quarterly GDP series available for

Mexico. The price of West Texas Intermediate crude oil, deflated by the CPI for

the United States, is used as an approximation of the world relative price of oil.

The sampling period extends from the second quarter of 1965 to the second

quarter of 1994. The real price of oil and the year-over-year growth rates of

industrial production and the monetary base are shown in Charts A-1 to A-3 of

Appendix 1.

13. A caveat to this is that the transitory component of potential output could be partlyaccounted for by the systematic response of fiscal and monetary policies to supply shocks.One could argue that the effect of this systematic response should not be included in thecalculation of potential output.

9

Our approach assumes the presence of a permanent (stochastic)

component in the level of output and the price of oil. Unit-root tests could not

reject these assumptions (see Appendix 2 for more details on these tests). The

same tests were also applied to the monetary base. Test results indicate that this

variable is stationary in first-difference form.

While unit-root tests suggest that the model variables are

nonstationary in levels, it is still possible that a stationary linear combination of

the variables could be found. In such a case, a vector error-correction model

would have to be estimated, since estimating a VAR in first differences would

remove important information about the behaviour of the variables that is

contained in the common trend. We used the method proposed by Johansen

(1988), and applied by Johansen and Juselius (1990), to test for cointegration

between the three variables (see Appendix 2 for more details). The tests could not

reject the null hypothesis of no cointegration. Consequently, we assume that the

series in the model are not cointegrated and that it is appropriate to estimate the

VAR models as a first difference (of the logarithms).

4 Results

This section includes four subsections. In the first, we present the variance

decomposition of Mexico’s output. We then present impulse responses of output to

demand, oil and supply shocks. In the third subsection, we present our estimates

of the cumulative effect of the different types of shocks on Mexico’s potential

output. Finally, we show how potential output can be projected using our

methodology.

10

4.1 Variance decomposition

The decomposition of variance presented in Table 1 allows us to measure the

relative importance of supply shocks, world oil shocks and demand shocks

underlying output fluctuations over different time horizons. Since we are

imposing the restriction that demand shocks have no permanent effect on output,

the proportion of output variance explained by supply and oil shocks gradually

approaches 100 per cent in the long run. Moreover, since all three types of shocks

are uncorrelated by assumption, the proportion of the output variance caused by

the sum of the three shocks is always equal to 100 per cent.

Table 1 suggests that demand shocks have played an important role

as a source of output fluctuation over very short horizons.14 However, the results

also illustrate how important oil shocks have been for the Mexican economy. In

effect, oil shocks appear to dominate from two quarters on. Non-oil supply shocks

have less importance over the period under consideration. A caveat to this

analysis is the uncertainty surrounding the numbers as illustrated by the

relatively wide 90 per cent confidence intervals. This is not surprising, since most

econometric studies present variance decompositions with very large confidence

intervals at conventional levels. The intrinsic volatility of Mexican economic series

and the large number of lags that are included in the VAR may accentuate the

problem in the context of our study.

14. Demand shocks are more important here than in Lalonde and St-Amant (1993). This canbe attributed to the method used by these authors to derive the number of lags for theirVAR, which led to the selection of too few lags and may have biased the estimation of thestructural components.

11

4.2 Impulse responses

Chart 1 (p. 12) shows the impulse response of Mexican output to the three

different kinds of structural shocks. Each shock is one standard deviation in size.

The horizontal axis is the number of years. As expected, world oil and supply

shocks have a permanent impact on the level of output, while most of the effects

of the demand shocks disappear after four years. Note that the impact of the

supply shock is felt only gradually, possibly reflecting the cost of adjustment to

these shocks. The chart, again, confirms the importance of world oil shocks for

the Mexican economy.

a. 90% confidence interval

TABLE 1: Variance decomposition of output(relative contribution of the different types of shocks, in per cent)

Horizon(quarters) World oil Supply Demand

1 27(2-64)a

18(0-78)

55(0-83)

2 44(7-78)

17(1-75)

39(2-68)

4 58(15-87)

12(1-75)

31(1-63)

8 63(16-88)

16(1-80)

21(2-55)

16 65(11-91)

24(2-81)

11(3-40)

32 67(7-93)

29(2-33)

5(2-33)

long term 67(5-96)

31(1-85)

0(0-35)

12

Chart 1: Response of Mexico’s index of industrial production

4.3 The components of output and potential output

Chart 2 shows the estimated stochastic supply components of Mexican output

(the cumulated effect of oil and supply shocks), expressed in a percentage of

actual output. This chart reaffirms the importance of oil for Mexico. In the

1970s, the effects of positive oil shocks, associated with the sharp increase in the

price of oil (see Chart A-1 in Appendix 1), had a major impact on the level of

Mexico’s output. The subsequent fall in oil prices accounts for the drop in the oil

component from 1982 to 1988. After some reversal in the late 1980s, the negative

oil shocks again adversely affected output growth in the 1990s, reflecting the

decline of the world price of oil.

Movements in the supply component are more difficult to interpret,

since they can be caused by many different kinds of shocks (demography,

productivity, and so forth). However, note the recent increase in the supply

13

component (it is less negative), which is compensating for the negative oil shock.

This increase could be due to the structural reforms that have been implemented

in recent years in Mexico, including the privatization that has occurred in some

sectors of the economy and the opening of its external sector in the context of the

North American Free Trade Agreement.

Chart 2: Supply components of output

Adding the world oil and supply components to the deterministic

component (that is, the trend) of output – or, alternatively, subtracting the

demand component from actual output – gives potential output, which is shown

on Chart 3 (p. 14) together with actual output.

14

Chart 3: Potential and actual output

The gap between actual and potential output, expressed as a

percentage of potential output, is what we define as the output gap, which is

shown in Chart 4 (p. 15). There is considerable uncertainty surrounding the

estimation of the output gap as illustrated by the fact that there are few periods

of statistically significant excess demand (actual output above potential) or

excess supply (negative gaps) when the conventional 90 per cent confidence

interval is used. This is why we also show a 67 per cent confidence interval.

Although the output gap fluctuates markedly (as does Mexican

output), there are substantial episodes of excess demand and excess supply. In

particular, note the persistent excess demand of the late 1970s and the early

1980s, which undoubtedly contributed to the acceleration of inflation during that

period (inflation is shown in Chart A-4 of Appendix 1). These excess demand gaps

may reflect the expansionary fiscal and monetary policies implemented in Mexico

after abandonment of the 1977 stabilization program of the International

15

Monetary Fund. After a period of excess supply, around 1983-85, large excess

demand reemerged around 1986-87. This again, probably reflects the easing of

fiscal and monetary policies. Inflation accelerated during those years. Finally,

note that since 1992 a large excess supply gap has appeared in Mexico. This can

be attributed to the tight financial policies followed during this period, which also

saw a deceleration of inflation.

Chart 4: Output gap(actual minus potential output in per cent of potential output)

4.4 Projecting potential output

We can project the level of Mexico’s potential output by adding the effect of past

supply shocks to the deterministic trend in industrial production. This is done in

Chart 5 (p. 16), which shows estimated and projected potential output for Mexico.

Note that projected potential output tends toward the estimated deterministic

trend in industrial production as the effect of past supply shocks subsides.

1970 1975 1980 1985 1990-8

-6

-4

-2

0

2

4

6

8

10

90%

75%

gap

16

Chart 5: Estimated and projected potential output

5 Conclusions

In this paper, we show how potential output can be estimated and projected using

an approach derived from the SVAR methodology, which has many advantages

over other approaches. Applying this method to the specific case of the Mexican

economy, we find that world oil shocks have been an important source of both

actual and potential output fluctuations over a sample period extending from

1965 to 1994. However, we also find occurrences of important gaps between

actual and potential output.

Because of limitations in the data on the Mexican economy, we

confined ourselves to a three-variable VAR representation model approach.

Models including more variables could be used to identify other types of shocks,

which could reduce the uncertainty surrounding the estimation of potential

projected

17

output and the output gap. We intend to look at such models to estimate potential

output in other countries.

19

Appendix 1Charts of the series

Chart A-1: Price of West Texas intermediate crude oil(deflated by the U.S. CPI)

Chart A-2: Industrial production(year-over-year growth rate)

20

Chart A-3: Monetary base(year-over-year growth rate)

Chart A-4: Consumer prices(year-over-year growth rate)

21

Appendix 2Unit-root and cointegration tests

Table A-1 shows the results of the augmented Dickey-Fuller (Dickey and

Fuller 1979), Phillips and Perron (1988) and Schmidt and Phillips (1992)

tests of the null hypothesis of nonstationarity of the price of oil, industrial

production and the monetary base. The results are unambiguous and

clearly support the hypothesis that these series are stationary in first

differences.

Table A-1: Unit-root tests(sample: 1965Q2-1994Q2)

a. The tests presented here allow for the presence of a linear trend in the series inlevel form, but assume that there is no deterministic trend in the differencedseries. The critical limits at a 5 per cent significance level of the ADF and the PPtests are 3.45 and 20.7, respectively, for the tests that include a linear trend, and2.89 and 13.7, respectively, for the tests that do not include a deterministic trend.The critical limit at a 5 per cent significance level of the SP test is 18.1. Boldfigures indicate that the unit-root hypothesis is rejected.

b. The number of lags for the ADF and SP tests was chosen using the recursiveprocedure suggested by Ng and Perron (1993).

c. The choice of the lag lengths for the PP test is related to the size of the sampleaccording to formulas suggested by Schwert (1989).

Series (in logarithms)Test statisticsa

ADFb PP (l=4)c PP (l=12) SPb

Industrial production (level) 1.65 4.18 3.92 3.31

Industrial production (first difference) 4.46 121.36 112.72 94.18

Monetary base (level) 2.19 3.22 3.79 0.44

Monetary base (first difference) 3.01 44.12 57.33 37.94

Oil prices (level) 0.76 2.69 3.01 3.80

Oil prices (first difference) 9.0 89.45 114.05 104.04

22

While unit root tests suggest that the variables are nonstationary in

levels, it is still possible that a stationary linear combination of these levels could

be found. To test for this, we used the method proposed by Johansen (1988) and

applied by Johansen and Juselius (1990). The results of the MV and TR tests

presented in Table A-1 support the hypothesis that there is no cointegrating

relationship between the variables considered in this paper. Note that we also

applied the single-equation procedure suggested by Engle and Granger (1987)

but could not find evidence of cointegration using that approach either (these

results are not presented here).

Table A-2: Cointegrating rank statistics(sample: 1965Q2 to 1994Q2)

MV TR

H0: r = 0 15.1 26.3

95% critical value 20.9 29.7

Note: The number of lags to be included in thevector error-correction model (10) wasselected according to a general-to-specificstrategy on the basis of an LR test usinga 10 per cent critical value.

23

Bibliography

Blanchard, O. J. and D. Quah. 1989. “The Dynamic Effect of Aggregate Demand and SupplyDisturbances.” American Economic Review 79: 655-73.

Blanchard, O. J. and M. W. Watson. 1986. “Are Business Cycles All Alike?” In The AmericanBusiness Cycle: Continuity and Change, edited by R. J. Gordon, 123-56. NBER Studies InBusiness Cycles, vol. 25. Chicago: University of Chicago Press.

Boschen, J. and L. Mills. 1990. “Monetary Policy with a New View of Potential GNP.” FederalReserve Bank of Philadelphia Business Review (July-August): 3-10.

DeSerres, A. and A. Guay. 1995. “The Selection of the Truncation Lag in Structural VARs (orVECMs) with Long-Run Restrictions.” Forthcoming working paper. Bank of Canada.

Dickey, D. A. and W. A. Fuller. 1979. “Distribution of the Estimator for Autoregressive TimeSeries with a Unit Root.” Journal of American Statistical Association 74: 427-31.

Engle, R. F. and C. W. J. Granger. 1987. “Cointegration and Error Correction: Representation,Estimation and Testing.” Econometrica 55 (March): 251-76.

Evans G. and L. Reichlin 1994. “Information, Forecasts, and Measurement of the BusinessCycle.” Journal of Monetary Economics 33: 233-54.

Jeong, J. and G. S. Maddala. 1993. “A Perspective on Application of Bootstrap Methods inEconometrics.” In Handbook of Statistics, vol. 11, edited by G. S. Maddala, C. R. Rao andH. D. Vinod, 573-610. New York: North-Holland.

Johansen, S. 1988. “Statistical Analysis of Cointegration Vectors.” Journal of EconomicDynamics and Control (June-September): 231-54.

Johansen, S. and K. Juselius. 1990. “Maximum Likelihood Estimation and Inference onCointegration – With Applications to the Demand for Money.” Oxford Bulletin ofEconomics and Statistics 51: 169-210.

King, R. G., G. H. Plosser, J. H. Stock, and M. W. Watson. 1991. “Stochastic Trends andEconomic Fluctuations.” American Economic Review 81 (September): 819-40.

Kuttner, K. N. 1994. “Estimating Potential Output as a Latent Variable.” Journal of Businessand Economic Statistics 12: 361-68.

Lalonde, R. and P. St-Amant. 1993. “Zones Monétaires Optimales : le Cas du Mexique et desÉtats-Unis.” Working Paper 93-12. Bank of Canada. Ottawa.

Laxton, D. and R. Tetlow. 1992. A Simple Multivariate Filter for the Measurement of PotentialOutput. Technical Report No. 59. Ottawa: Bank of Canada.

Ng, S. and P. Perron. 1993. “Unit Root Tests in Autoregressive-Moving Average Models withData Dependent Methods for the Truncation Lag.” Mimeo. Université de Montréal.

24

Osterwald-Lenum, M. 1992. “A Note with Quantiles of the Asymptotic Distribution of theMaximum Likelihood Cointegration Rank Test Statistics.” Oxford Bulletin of Economicsand Statistics 54: 461-72.

Phillips, P. C. B. and P. Perron. 1988. “Testing for a Unit Root in Time Series Regressions.”Biometrika 75: 335-46.

Pitarakis, J.Y. 1993. “Cointegration Analysis in Large Systems: New Test Criteria and aModel Selection Procedure.” Mimeo. Department of Economics, Boston University.

Reinsel, G. C. and S. H. Ahn. 1988. “Asymptotic Distribution of the Likelihood Ratio Test forCointegration in the Nonstationary VAR Model.” Technical Report University ofWisconsin, Madison.

Runkle, D. E. 1987. “Vector Auto-regression and Reality.” Journal of Business and EconomicStatistics 5: 437-54.

Schwert, G. W. 1989. “Tests for Unit Roots: A Monte Carlo Investigation.” Journal of BusinessEconomic Statistics 7: 147-59.

Sims, C. A. 1986. “Are Forecasting Models Usable for Policy Analysis?” Federal Reserve Bankof Minneapolis Quarterly Review 10 (Winter): 2-16.

Shapiro, M. D. and M. W. Watson. 1988. “Sources of Business Cycle Fluctuations.” WorkingPaper 2589. National Bureau of Economic Research, Cambridge (MA).

Watson, M. W. 1986. “Univariate Detrending Methods with Stochastic Trends.” Journal ofMonetary Economics 18: 49-75.

Zivot, E. and D. W. K. Andrews. 1992. “Further Evidence on the Great Crash, the Oil-PriceShock, and the Unit-Root Hypothesis.” Journal of Business & Economic Statistics 10:251-70.

Bank of Canada Working Papers

1995

95-1 Deriving Agents’ Inflation Forecasts from the Term Structure of Interest Rates C. Ragan

95-2 Estimating and Projecting Potential Output Using Structural VAR A. DeSerres, A. GuayMethodology: The Case of the Mexican Economy and P. St-Amant

1994

94-1 Optimum Currency Areas and Shock Asymmetry: N. Chamie, A. DeSerresA Comparison of Europe and the United States and R. Lalonde

94-2 A Further Analysis of Exchange Rate Targeting in Canada R. A. Amano and T. S. Wirjanto

94-3 The Term Structure and Real Activity in Canada B. Cozier and G. Tkacz

94-4 An Up-to-Date and Improved BVAR Model D. Racette, J. Raynauldof the Canadian Economy and C. Sigouin

94-5 Exchange Rate Volatility and Trade: A Survey A. Côté

94-6 The Dynamic Behaviour of Canadian Imports and the Linear- R. A. AmanoQuadratic Model: Evidence Based on the Euler Equation and T. S. Wirjanto

94-7 L’endettement du secteur privé au Canada : un examen macroéconomique J.-F. Fillion

94-8 An Empirical Investigation into Government Spending R. A. Amanoand Private Sector Behaviour and T. S. Wirjanto

94-9 Symétrie des chocs touchant les régions canadiennes A. DeSerreset choix d’un régime de change and R. Lalonde

94-10 Les provinces canadiennes et la convergence : une évaluation empirique M. Lefebvre

94-11 The Causes of Unemployment in Canada: A Review of the Evidence S. S. Poloz

94-12 Searching for the Liquidity Effect in Canada B. Fung and R. Gupta

1993(Earlier 1993 papers, not listed here, are also available.)

93-15 Oil Prices and the Rise and Fall of the U.S. Real Exchange Rate R. A. Amano and S. van Norden

Single copies of Bank of Canada papers may be obtained from Publications DistributionBank of Canada234 Wellington StreetOttawa, Ontario K1A 0G9